Tipton_unc_0153M_17156.pdf

97 

Full text

(1)

EVALUATION OF A LOW-COST COMPARTMENT BAG TEST TO QUANTIFY HYDROGEN SULFIDE-PRODUCING BACTERIA IN DRINKING WATER

Claire Carter Tipton

A thesis submitted to the faculty at the University of North Carolina at Chapel Hill in partial

fulfillment of the requirements for the degree of Master of Science in the Department of

Environmental Sciences & Engineering in The Gillings School of Global Public Health.

Chapel Hill

2017

Approved by:

Jill R. Stewart

Mark D. Sobsey

(2)
(3)

iii ABSTRACT

Claire Carter Tipton: Evaluation of a low-cost compartment bag test to quantify hydrogen sulfide-producing bacteria in drinking water

(Under the direction of Jill Stewart)

Tests for detecting hydrogen sulfide (H2S)-producing bacteria as fecal indicators have been proposed to assess drinking water safety in low-resource settings. This study compared a semi-quantitative compartment bag test (CBT) to the EPA- and FDA-approved multiple test tube (MTT) method to quantify H2S-producing bacteria in drinking water sources. Both methods used PathoScreen™ medium to detect target bacteria in 60 surface water samples collected from North Carolina drinking water reservoirs. Samples were subjected to paired levels of incubation temperatures (25° C, 35° C) and numbers of incubation days (1, 2, 3). Results indicated a significant positive correlation between

(4)

iv

To my mentor and friend, I couldn’t have done this without you.

(5)

v

ACKNOWLEDGEMENTS

I would like to acknowledge the following people who made this study possible:

First, I would like to thank my undergraduate research assistants, whose enthusiasm and dedication made field and lab work flow– Ryan Leighton, Rachel Lempp, Lindsey Wickersham and Joseph Maitre. I would also like to thank everyone in the Stewart lab, especially Elizabeth Christenson and Sarah Rhodes, for their guidance and assistance during the lab process. I would also like to thank Emily Bailey and everyone in the Sobsey lab for graciously sharing lab equipment and techniques.

Many thanks to statistician Dillion Ashburn for partnering with me to analyze and interpret my data. I’d especially like to thank Dr. Matt Psioda with UNC Biostatistics and Dr. Andrew Gronewold with NOAA for their subject matter expertise and feedback on statistical theory and processes.

I would like to acknowledge the tremendous support of my thesis committee: Dr. Jill Stewart, Dr. Mark Sobsey, and Dr. Stephen Whalen. Thank you Dr. Sobsey for your subject matter expertise, feedback, and openness to new ideas. I am extremely grateful for my research advisor and mentor, Dr. Jill Stewart. Dr. Stewart, thank you so much for your insight, feedback, and ceaseless support.

I would like to thank all the staff at Aquagenx, LLC for their partnership, collaboration, and support of the project. A special thank you goes to company liaison Lisa Hirsh– it has been a pleasure working with you. I would also like to thank the Orange Country Water and Sewer Authority for graciously allowing us to sample from their drinking water reservoirs.

(6)

vi

TABLE OF CONTENTS

LIST OF TABLES ... vii

LIST OF FIGURES ... viii

LIST OF ABBREVIATIONS AND SYMBOLS ... ix

CHAPTER 1: INTRODUCTION ... 1

CHAPTER 2: OBJECTIVES ... 5

CHAPTER 3: LITERATURE REVIEW ... 6

CHAPTER 4: METHODS ... 23

CHAPTER 5: RESULTS ... 33

CHAPTER 6: DISCUSSION ... 54

CHAPTER 7: CONCLUSION ... 67

APPENDIX I: CBT REFERENCE RESULT CHART ... 69

APPENDIX II: MTT REFERENCE RESULT CHART ... 70

APPENDIX III: ALL STUDY CBT AND MTT OUTCOMES ... 71

APPENDIX IV: COMPLETE DATASET ... 73

(7)

vii LIST OF TABLES

Table 1: Number of samples by dilution and incubation temperature ... 27

Table 2: Correlation coefficients of CBT vs. MTT outcomes ... 35

Table 3: Comparison of CBT vs. MTT descriptive statistics (overall) ... 39

Table 4: Comparison of CBT vs. MTT descriptive statistics (by incubation temperature and days) ... 39

Table 5: Comparison of CBT vs. MTT descriptive statistics (when outcomes ≤ 100 MPN/100 mL) ... 40

Table 6: Contingency table of CBT vs. MTT presence/absence outcomes (by incubation days) ... 43

Table 7: CBT sensitivity, specificity, FPR, FNR, PPV, NPV, and accuracy calculations ... 43

Table 8: Contingency table of CBT vs. MTT presence/absence outcomes (by sample) ... 44

Table 9: CBT sensitivity, specificity, FPR, FNR, PPV, NPV, and accuracy calculations ... 44

Table 10: Contingency table of CBT vs. MTT by day (1-10 MPN/100 mL and 11-100 MPN/100 mL) ... 45

Table 11: Contingency table of CBT vs. MTT by sample (1-10 MPN/100 mL and 11-100 MPN/100 mL) ... 45

Table 12: Contingency table of CBT vs. MTT by WHO health risk categories ... 46

Table 13: Frequency counts of CBT results falling within MTT 95% confidence intervals... 47

Table 14: Linear regression of CBT vs. MTT outcomes (by incubation temperature and days) ... 48

Table 15: Correlation coefficients of CBT outcomes by different incubation conditions ... 50

(8)

viii LIST OF FIGURES

Figure 1: Map of sampling locations at University Lake. ... 24

Figure 2: Map of sampling locations at Cane Creek Reservoir………..25

Figure 3: Formula of Thomas (1942) equation ... 28

Figure 4: Part I of Method of Haldane (1939) equation ... 29

Figure 5: Part II of Method of Haldane (1939) equation ... 29

Figure 6: Scatter plot of the highest correlation between CBT vs. MTT outcomes ... 35

Figure 7: Scatter plot of high correlation between CBT vs. MTT outcomes. ... 36

Figure 8: Scatter plot of high correlation between CBT vs. MTT outcomes ………37

Figure 9: Scatter plot worst correlation between CBT vs. MTT outcomes ... 38

Figure 10: Box & whisker plots of CBT vs. MTT outcomes (overall) ... 40

Figure 11: Box & whisker plots of CBT vs. MTT outcomes (by incubation temperature) ... 41

Figure 12: Box & whisker plots of CBT vs. MTT outcomes (by numbers of incubation days)... 42

Figure 13: Linear regression hypothesis test ... 47

(9)

ix

LIST OF ABBREVIATIONS AND SYMBOLS ° C Degrees Celsius

CBT Compartment bag test CC Cane Creek Reservoir CI Confidence interval

DO Dissolved oxygen

E. coli Escherichia coli

EPA United States Environmental Protection Agency ° F Degrees Fahrenheit

FC Fecal coliform

FDA United Stated Food and Drug Administration FeS Ferrous sulfide

FIB Fecal indicator bacteria FIO Fecal indicator organisms FNR False negative rate FPR False positive rate H2S Hydrogen Sulfide

ISO International Organization for Standardization

Ln Natural log

Log10 Log base 10

MDG Millennium Development Goals mg/L Milligrams per liter

mL Milliliter

(10)

x MTT Multiple test tube method

NGO Non-governmental organizations NPV Negative predictive value

OR Odds Ratios

P P-value

PA Presence/Absence

PBS Phosphate Buffered Solution pH Potential of Hydrogen PPV Positive predictive value

r Spearman's rho

r2 Coefficient of determination SRB Sulfate reducing bacteria

TC Total coliform

TRFLP Terminal Restriction Fragment Length Polymorphism

UL University Lake

UNICEF United Nations International Children’s Fund

USD United States Dollar

USAID United States Agency for International Development W Shapiro-Wilk normality test statistic

(11)

1

CHAPTER 1: INTRODUCTION

Safe drinking water is a fundamental human right and requirement for good health. Despite this, fecal contamination in drinking water affects over 1.8 billion people worldwide and is estimated to cause over 500,000 diarrheal deaths each year. Young children are particularly vulnerable, with approximately 361,000 diarrheal deaths occurring each year to children under five due to unsafe drinking water conditions (WHO 2016). Many of the world’s population impacted by poor microbial

water quality reside in rural or low-resource environments (Anwar et al. 1999). While many technologies exist to detect fecal contamination in drinking water, the high level of human, financial, and

technological capacity required to conduct such tests poses logistical challenges to routine monitoring in these settings (Crocker & Bartram 2014). Similar challenges arise during humanitarian emergencies such as natural disasters and wartime conflict, which often jeopardize utilities and drinking water supplies (Adams 1999).

To overcome this problem, field tests that detect hydrogen sulfide (H2S)-producing bacteria as indicators of fecal contamination in water have been proposed for use in low-resource settings. The H2 S-producing bacteria field test (H2S test) has been compared to traditional methods for detecting fecal indicator bacteria (FIB), and has demonstrated relatively good correlation with conventional indicator organisms (Ratto et al. 1989; Castillo et al. 1994; Venkobachar et al. 1994; Genthe & Franck 1999; Rijal & Fujioka 2001; Manja et al. 2001; Sobsey & Pfaender 2002; McMahan 2011; Tambi et al. 2016;

(12)

2

including simple format, low-cost of materials, and ease of isolating, identifying, and enumerating target organisms (McMahan 2011).

The H2S test is based on the reaction of iron in the medium with hydrogen sulfide gas produced by sulfate reducing bacteria (SRB), causing the formation of the black insoluble precipitate ferrous sulfide. H2S-producing bacteria are found in a wide variety of habitats, including freshwater, and include many enteric pathogens such as Salmonella, Proteus, Edwardsiella, Yersinia, Aeromonas, Clostridium, Staphylococcus, Peptococcus, and Campylobacter, fecal coliforms (Citrobacter freundii, Klebsiella pneumoniea, K. oxytoca, Enterobacter cloacae), andH2S-producing variants of Escherichia coli (E. coli). Although widely distributed in the environment, H2S-producing bacteria have been found to be

consistently associated with fecal contamination (McMahan 2011; Manja et al. 1982; Manja et al. 2001; Nagaraju & Sastri 1999; Venkobachar et al. 1994; Nair et al. 2001; Ratto et al. 1989; Kaspar et al. 1992; Castillo et al. 1994; Martins et al. 1997; Kromoredjo & Fujioka 1991; Genthe & Franck 1999; Sivaborvon 1988; Sobsey & Pfaender 2002; McMahan et al. 2012). They also meet other criteria required for ideal fecal indicator bacteria (FIB), such as similar survival and transport in the environment compared to pathogens (Martins et al. 1997; Castillo et al. 1994; Nagaraju & Sastri 1999), present in greater numbers than pathogens (Nagaraju & Sastri 1999; Castillo et al. 1994; Manja et al. 1982), broad applicability (Ratto et al. 1989; Castillo et al. 1994; Martins et al. 1997; Kromoredjo & Fujioka 1991; Genthe & Franck 1999; Sivaborvon 1988), quantifiable (McMahan 2011; McMahan et al. 2012; Venkobachar et al. 1994; Rijal & Fujioka 2001; Manja et al. 2001), adequate sensitivity (Ratto et al. 1989; Anwar et al. 1999; Genthe & Franck 1999; Rijal & Fujioka 2001; Roser et al. 2005), and logistic feasibility (Genthe & Franck 1999; Bain et al. 2012; Mosley & Sharp 2005; Venkobachar et al. 1994; Castillo et al. 1994; Ratto et al. 1989; Anwar et al. 1999; S P Pathak & Gopal 2005; Hirulkar & Tambekar 2006; Khush et al. 2013;

(13)

3

is needed, studies by McMahan (2011) and Kush et al. (2013) show potential for the H2S test to be associated with risk of diarrheal illness.

The H2S field test was first developed in the early 1980’s as a simple and reliable

presence/absence (PA) test for village health workers in India (Manja et al. 1982). Since then, many modifications to the test have been developed, including changes to medium composition, medium preparation, sample volume, incubation times and temperatures, test formats, and methods to score results. The H2S test, particularly PA versions of the test, has received widespread use and

commercialization in recent years (Bain et al. 2012).

While the H2S test has been used globally for over three decades, the method is still under debate among scientists and regulatory agencies. One of the main concerns regarding the test is the lack of standardization across lab- and commercially-made tests (Sobsey & Pfaender 2002). The proliferation of presence/absence tests vs. more quantitative tests is also a concern, especially in light of the World

Health Organization’s (WHO) Guidelines for Drinking Water Quality shifting towards quantifiable risk-based data (Bain et al. 2012). Differing sensitivity and specificity results among studies has also raised concerns (Wright et al. 2012; Izadi et al. 2010; Tewari et al. 2003; Desmarchelier et al. 1992; Yang et al. 2013a). Until more rigorous research can be done to generate method consensus, as well as provide further evidence to support H2S-producing bacteria as a viable FIB, regulatory agencies such as the United States Environmental Protection Agency (EPA) and the WHO will not accept the H2S test for microbial water quality purposes.

(14)

4

Morrison 2016; Gerges et al. 2016; McMahan et al. 2017). The kit is currently manufactured and distributed by Aquagenx, LLC (Chapel Hill, NC, USA).

In an effort to further validate H2S-producing bacteria as alternative fecal indicator organisms, McManhan (2011) assessed the feasibility of a combined H2S compartment bag test (H2S CBT) using lab-made and proprietary H2S substrates. Researchers ran a cost analysis of the H2S CBT method against other common microbial water quality tests, including MI Agar, BioRad Rapid E. coli 2 Agar, Coliert, Petrifilm E. coli, EasyGel, and the E. coli CBT. The H2S CBT was judged to be the most cost-effective, at $0.40 per sample (McMahan 2011). While results were promising, no further work has been done to evaluate and compare the capabilities of the H2S CBT to standard semi-quantitative methods.

The purpose of this study was to compare the compartment bag test to the EPA- and United States Food and Drug Administration (FDA)-approved multiple test tube (MTT) method to enumerate H2S-producing bacteria in drinking water sources using a most probable number (MPN) format. Development of an inexpensive, simple, and semi-quantitative H2S field test would provide more quantitative information on human health risk related to microbial water quality than more common PA H2S tests. Validation of a reliable and semi-quantitative H2S test would also aid in the effort to

(15)

5

CHAPTER 2: OBJECTIVES There were two primary objectives for this study:

1. Compare the compartment bag test (CBT) to the multiple test tube (MTT) method to detect and quantify H2S-producing bacteria in drinking water sources using a most probable number (MPN) format.

(16)

6

CHAPTER 3: LITERATURE REVIEW Introduction

Fecal contamination of drinking water is a major cause of waterborne illnesses in humans worldwide, with microbial contamination responsible for the great majority of the water-related health burden (WHO 2008). Since the early 1980’s, methods for detecting fecal contamination using hydrogen sulfide (H2S)-producing bacteria as alternative fecal indicator bacteria (FIB) have been developed (Manja et al. 1982). The H2S-producing bacteria field test (H2S test) has been advocated for use in low-resource and humanitarian emergency settings due to its low-cost and user friendly format. Additionally, there is a growing body of evidence to support H2S-producing bacteria as credible

alternative fecal indicator organisms.

One of the challenges facing the H2S test is the lack of standardization across the method. Multiple versions of the H2S test exist, with differences in medium composition, sample volumes, test formats, and methods to score results limiting the ability to validate and compare the H2S test to more traditional FIB methods (Sobsey & Pfaender 2002). In addition, there are few semi-quantitative or quantitative H2S tests on the market compared to presence/absence (PA) tests. PA tests cannot determine microbial concentrations nor estimate health risks related to microbial water quality, which are important factors for determining water safety in resource-limited environments.

(17)

7

confirmed (McMahan 2011), more comparisons of the test to standard microbial water quality methods that estimate bacteria concentrations by quantal methods are needed.

The goal of this review is to discuss the published literature on H2S field tests, including the usefulness of H2S-producing bacteria as indicators of fecal contamination in low-resource settings, the history of method development and subsequent modifications, and the introduction and initial validation of the compartment bag test as a means of quantifying target bacteria in water. Fecal contamination of drinking water and diarrheal illness

Fecal contamination in drinking water impacts over 25% of the world’s population and is estimated to cause over 500,000 diarrheal deaths each year (WHO 2016). Diarrheal disease is a major cause of morbidity and mortality in all age groups, but especially among young children. For instance, unsafe drinking water conditions are estimated to cause over 360,000 diarrheal deaths ever year in children under five (WHO 2016). While mortality from diarrheal illnesses has decreased over the past fifty years, a study conducted in 2003 suggests there has not been an accompanying decrease in morbidity on the global burden of disease (Kosek et al. 2003).

Water is one of the primary pathways for transmission of diarrheal illnesses. Transmission of disease via water can be classified into four categories: waterborne, water-washed, water-based, and water-related (White et al. 2002). Waterborne diseases occur when pathogens are ingested via the fecal-oral route and are the source of illnesses such as gastroenteritis, giardia, cholera, and infectious hepatitis (Cairncross & Feachem 1993). Waterborne pathogens comprise a broad range of

(18)

8 Fecal indicator organisms

Detecting and monitoring fecal pathogens in drinking water is crucial to managing water systems and minimizing disease risk as well as protecting local and global public health. However, methods to detect the full spectrum of pathogens that may occur in water are currently cost prohibitive and impractical to implement on a widespread scale (US EPA 2009). For decades, regulatory agencies and scientific governing bodies around the world have promoted the use of fecal indicator organisms as surrogates for potential pathogens and subsequent health risks in recreational and drinking water sources (US EPA 2009). Fecal indicator organisms (FIO) are microorganisms found in the intestines of warm-blooded animals, including humans, and are shed in feces. While FIO are generally not hazardous to human health, their presence and density in water indicate the possible presence of pathogenic organisms due to fecal contamination. To qualify as an FIO, candidate microbes and the tests that detect them ideally meet the following criteria according to The Routledge Handbook of Water and Health (Bartram et al. 2015):

• Be present whenever enteric pathogens are present

• Be absent whenever enteric pathogens are absent, or at levels that pose no increased risk

• Be present in greater numbers than pathogens

• Have similar or greater survival rates than pathogens in the environment

• Have broad applicably and detectability in all types of water that humans may encounter

• Be specific to a fecal source with humans or species who share fecal-oral pathogens with humans

• Do not multiply independently in the environment

• Be reliably, rapidly, and distinctly detectable at low-cost

(19)

9

While no fecal indicator organisms to date satisfy all requirements under all circumstances, many regulatory agencies and scientific governing bodies consider Escherichia coli (E. coli), enterococci,

and members of the fecal coliform group as the “gold standards” of microbial water quality testing (US EPA 2009). Total coliforms, fecal streptococci, Clostridium perfringens, and coliphages are also fecal indicator organisms.

Many presence/absence (PA), semi-quantitative, and quantitative methods have been developed to detect FIO in water resources. PA tests provide simple positive-negative results and are most applicable in situations where water is usually uncontaminated and when most samples provide negative test results. On the other hand, semi-quantitative and quantitative methods provide both positive-negative results and microbial concentration estimations. These features are useful for

categorizing water safety levels and estimating potential human health risk. Methods of this nature are most applicable in situations where fecal contamination is likely (WHO 1996). Furthermore, quantitative methods are often desired to satisfy the information and monitoring needs of operational, compliance, and surveillance sampling regimes (Bain et al. 2012). Widely recognized microbial water quality methods approved by organizations such as the United States Environmental Protection Agency (EPA), the World Health Organization (WHO), and the International Organization for Standardization (ISO) include most probable number (MPN), membrane filtration, quantitative polymerase chain reaction (qPCR), and use of defined substrates such as the Coliert Quanti-Tray MPN test (Fewtrell & Bartram, 2001; U.S. EPA, 2015).

Challenges monitoring water quality in low-resource settings

(20)

10

environments (Anwar et al. 1999). Furthermore, acute diarrheal infections are one of the most frequent childhood illnesses and reasons for visits at health clinics in low- and middle-income countries (Walker et al. 2013). The lack of human, financial, and technological capacity in low-resource settings limits the ability of countries and communities to monitor water resources (Crocker & Bartram 2014). Similar obstacles arise during humanitarian emergencies such as natural disasters or wartime conflict, which often put utilities and local water supplies in jeopardy. Safe water and sanitation, along with food and shelter, receive the highest priority as first-phase interventions during emergency situations (UNICEF 2012). Due to logistical challenges frequently encountered in emergency and low-resource settings, the need for inexpensive, rapid, and reliable fecal contamination detection methods is paramount.

H2S-producing bacteria and the H2Sfield test

(21)

11

non-soluble black precipitate that is readily distinguishable and denotes a positive reaction (Madigan et al. 2008).

Overview of H2Stest history, method development, and subsequent modifications

In the initial 1982 report, Manja et al. compared a novel H2S presence/absence (PA) paper strip test to a standard E. coli MPN test to detect fecal contamination in drinking water samples in several cities in India. Drinking water samples were added to sterilized bottles containing a reagent of ferrous iron, sulfate salts, and nutrients to promote the growth and metabolism of the bacteria of interest. Samples were observed for black color change over 12-18 h and 24-48 h periods at ambient

temperatures (30-37° C). When E. coli was detected at levels greater than or equal to 10 MPN/100 mL, the sample was also tested using the H2S test. Researchers observed the presence of coliform bacteria in drinking water was consistently associated with organisms that produced H2S. They also reported good agreement between the two methods at higher levels of E. coli contamination (> 40 MPN/100 mL) (Manja et al. 1982).

Over the past three decades, many versions and modifications of the H2S test have been described in the literature. Modifications to the H2S test include changes in medium composition, preparation of the medium and supporting materials, test format, sample volumes, incubation time, incubation temperature, and scoring of results. Many investigators have evaluated the H2S method by comparing it to traditional fecal indicator bacteria (FIB) methods under controlled lab conditions or in tropical and subtropical regions such as Indonesia, Peru, Paraguay, Chile, Nepal, and South Africa (Ratto et al. 1989; Kromoredjo & Fujioka 1991; Kaspar et al. 1992; Castillo et al. 1994; Venkobachar et al. 1994; Rijal & Fujioka 2001; Genthe & Franck 1999).

(22)

12

Pathak & Gopal 2005; Shahryari et al. 2014; Sobsey & Pfaender 2002). Additionally, longer incubation periods between 24 h to 48 h and incubation temperatures in the range of 25-35° C have shown best results in terms of detecting low levels (5 CFU per sample) of H2S-producing bacteria (Pillai et al. 1999; Gawthorne et al. 1996; Ratto et al. 1989; Castillo et al. 1994; Genthe & Franck 1999; Manja et al. 2001; Tambekar & Neware 2012; Gupta et al. 2007; Sobsey & Pfaender 2002). The H2S presence/absence test format has been evaluated extensively, with many versions experiencing widespread commercialization and field use (Bain et al. 2012). More recently, semi-quantitative methods using MPN or membrane filtration formats have been developed (Venkobachar et al. 1994; Rijal & Fujioka 2001; McMahan 2011; Roser et al. 2005; McMahan et al. 2011; McMahan et al. 2012). The H2S test is typically performed using 10-100 mL sample volumes. Methods to score results typically employ presence/absence (positive-negative) results or MPN or CFU per sample volume concentration estimations. The H2S test has been frequently evaluated via comparison with traditional FIO including E. coli, fecal coliforms, total coliforms, fecal streptococci, and enterococci (Pillai et al. 1999; Ratto et al. 1989; Venkobachar et al. 1994;

(23)

13 Strengths of the H2Sfield test

Several investigators have attempted to determine the reliability of the H2S test for the detection of fecal contamination in drinking water. Overall, their research indicates there is a strong correlation between the H2S test and traditional fecal indicator bacteria, and that the H2S method detects fecally contaminated water with about the same frequency and magnitude as traditional comparison methods (Ratto et al. 1989; Castillo et al. 1994; Venkobachar et al. 1994; Genthe & Franck 1999; Rijal & Fujioka 2001; Manja et al. 2001; Sobsey & Pfaender 2002; McMahan 2011; Tambi et al. 2016; Sivaborvon 1988; Martins et al. 1997; Dufour et al. 2013; Manja et al. 1982; Anwar et al. 1999; Nair et al. 2001; Hirulkar & Tambekar 2006; Gupta et al. 2007; Eun & Hwang 2003; Kromoredjo & Fujioka 1991; McMahan et al. 2012).

Additionally, multiple studies have demonstrated the H2S test’s ability to meet many of the criteria required for consideration as an ideal fecal indicator. Similar or greater survival of H2S-producing organisms to pathogens has been demonstrated (Martins et al. 1997; Castillo et al. 1994; Nagaraju & Sastri 1999), along with greater numbers of H2S-producing bacteria than pathogens found in sample waters (Nagaraju & Sastri 1999; Castillo et al. 1994; Manja et al. 1982). While the H2S test does not consistently measure the presence of total coliforms, fecal coliforms, or E. coli, many members of the fecal coliform group are known H2S-producers, including Klebsiella pneumoniae, K. oxytoca,

Enterobacter cloacae, and Citrobacter freundii (LeClerc et al. 2001). Moreover, Sobsey and Pfaender (2002) suggest that organisms-producing positive H2S results may not all be coliforms but are typically associated with the intestinal tracts of warm-blooded animals. Castillo et al. (1994) found a large variety of bacteria in samples giving positive reactions in the H2S test, primarily Clostridium perfringens,

(24)

14

found H2S-producing bacteria to be specific to a fecal source or identifiable as to a source of origin via comparison to standard fecal indicators (McMahan 2011; Manja et al. 1982; Manja et al. 2001; Nagaraju & Sastri 1999; Venkobachar et al. 1994; Nair et al. 2001; Ratto et al. 1989; Kaspar et al. 1992; Castillo et al. 1994; Martins et al. 1997; Kromoredjo & Fujioka 1991; Genthe & Franck 1999; Sivaborvon 1988). The H2S test demonstrates broad applicability, as it has been applied to diverse global water sources

including groundwater, surface water, bore wells, dug wells, rainwater cisterns, and municipal water supplies (Ratto et al. 1989; Castillo et al. 1994; Martins et al. 1997; Kromoredjo & Fujioka 1991; Genthe & Franck 1999; Sivaborvon 1988). Adequate or superior detectability has been documented in many cases (Ratto et al. 1989; Anwar et al. 1999; Genthe & Franck 1999; Rijal & Fujioka 2001; Roser et al. 2005), and results can be rapidly obtained in 24 h for heavy-to-moderate contamination and 48 h for light contamination (Manja et al. 2001; Manja et al. 1982; Nagaraju & Sastri 1999; Venkobachar et al. 1994; Castillo et al. 1994; Martins et al. 1997; Genthe & Franck 1999; Rijal & Fujioka 2001; Weppelmann et al. 2014a; Izadi et al. 2010). While PA H2S tests do not provide quantifiable results, recently

developed H2S MPN tests have shown similar detection and agreement with standard FIB methods (McMahan 2011; McMahan et al. 2012; Venkobachar et al. 1994; Rijal & Fujioka 2001; Manja et al. 2001). Precision of results among samples has been documented, though not between labs (Martins et al. 1997; Rijal & Fujioka 2001). Genthe and Franck (1999) were also able to demonstrate measures of viability and infectivity with H2S-producing bacteria. While more research is needed, studies by McMahan (2011) and Kush et al. (2013) show potential for the H2S test to be associated with risk of diarrheal illness.

(25)

15

of Clostridium perfringens and related sulfite-reducing clostridia, which serve as better indicators of protozoan parasites such as Giardia cysts and Cryptosporidium oocysts (McMahan 2011). Roser et al (2005) also contended that the H2S test appears much more sensitive than measurements of somatic and male-specific (F+) coliphages and protozoan pathogens in their study, and that overall the H2S test shows fairly high sensitivity, specificity and precision when comparing results across studies (Roser et al. 2005).

Perhaps one of the most promising aspects of the H2S test is its practicality for low-resource and emergency settings. H2S test kits are relatively easy to manufacture and are often made locally at lower cost than standard methods (Genthe & Franck 1999). Bain et al. (2012) estimated the cost per test of four common commercialized H2S PA tests to range from $0.60 to $2.40 USD per sample. The H2S test has been applied in many developing countries, in emergencies (Mosley & Sharp 2005), and in remote areas of developed countries (UNICEF 2008). The test’s simple training and personnel needs, utility in the field, low-cost, and moderate volume requirements are reasons researchers have justified continued evaluation of the method (Genthe & Franck 1999; Venkobachar et al. 1994; Castillo et al. 1994; Ratto et al. 1989; Anwar et al. 1999; S. P. Pathak & Gopal 2005; Hirulkar & Tambekar 2006; Bain et al. 2012; Khush et al. 2013; Weppelmann et al. 2014a; Walker et al. 2013; Tambi et al. 2016; Sivaborvon 1988; Mosley & Sharp 2005), despite notable test weaknesses.

Weaknesses of the H2Sfield test

(26)

16

sensitivity have also been documented among labs (Wright et al. 2012; Izadi et al. 2010; Tewari et al. 2003; Desmarchelier et al. 1992; Yang et al. 2013a). Ideal fecal indicators should also be absent in unpolluted water and present only when waters are fecally contaminated. However, studies have shown that multiple microorganisms produce H2S, many of which occur naturally in waters that are not fecally contaminated (Sobsey & Pfaender 2002; Ratto et al. 1989; Kaspar et al. 1992; Venkobachar et al. 1994; Sivaborvon 1988; Martins et al. 1997). Similar concerns arise when the H2S test is conducted in waters with higher levels of naturally occurring iron or sulfide, potentially leading to false positive results. Previous studies applying the H2S test to groundwater samples have demonstrated false positive results, where H2S positive samples contained no fecal coliforms or E. coli (Kaspar et al. 1992; Pant et al. 2002). In this case, the rapid reaction of iron with sulfide already present in water samples could produce a darkening of H2S tests almost immediately upon addition of samples. For this reason, Sobsey & Pfaender (2002) advise visual inspection of H2S tests for quick or early positive reactions, between a few minutes to an hour of incubation.

While there appears to be no reasonable way to preclude all non-fecal H2S producers from water sources, understanding the ecology of H2S producers (i.e. sulfate reducing bacteria, SRB) may explain the likelihood of false positives of this nature. Wetzel (2001) noted that there would be little sulfate for bacteria to use if concentrations are low in freshwater (Wetzel 2001). On the other hand, in settings where sulfate concentrations are high, such as in geothermal environments, SRB could give false positive results in H2S tests. Another point to consider is that sulfate reducers do not metabolize

(27)

17

(Widdel 1988). However, for a positive reaction to occur, the test sample would need to become anaerobic, allowing the fermentative bacteria to produce the required short-chain organic acids and other preferred substrates leading to the growth of SRB in a test sample. These conditions are not as likely to be achieved in the incubation times typically used in H2S tests (1-2 days), though they are possible (Sobsey & Pfaender 2002). Furthermore, McMahan et al. (2012) demonstrated that a semi-quantitative H2S compartment bag test was not impacted by high sulfur and high iron levels in well-water samples. In addition, they demonstrated a consistent association between positive H2S test results and species identified in positive samples with fecal indicator organisms and enteric pathogens present in natural waters (lake, wells, and cistern rainwater) in the United States (McMahan et al. 2012).

Another major concern regarding the H2S test is the lack of standardization across tests. As previously described, numerous modifications and versions of the H2S test have been developed. Variations include medium composition, medium preparation (dried at elevated temperature, lyophilized, autoclaved only, etc.) sample volume (20 mL, 100 mL, etc.), paper use, paper type, and paper size to which the medium is absorbed, incubation times and temperatures, test formats (PA, semi-quantitative MPN, membrane filter enumeration), and methods to score results (Sobsey & Pfaender 2002). The multitude of different H2S test versions, as well as the variety of ways they have been evaluated in field and lab studies, makes comparisons across tests difficult (Wright et al. 2012) . While efforts have been made in India and the United States to make commercially prepared medium and implement performance criteria, the test is not standard worldwide and there has been no effort to achieve a standard test procedure (Sobsey & Pfaender 2002).

(28)

18

the absence of a microbial risk data for H2S PA tests raises concerns about their validity and interpretation in judging drinking water quality (Bain et al. 2012). This shift increases the need for development of a reliable and affordable quantitative H2S method. Semi-quantitative test formats provide more information than standard PA tests, as they provide a concentration estimate of H2 S-producing organisms in a given water sample. Having quantified or semi-quantified levels of fecal contamination is important for efforts to relate microbial contamination in water to waterborne disease risk. Semi-quantitative H2S tests can also be more easily compared to standard semi-quantitative fecal indicator methods, such as the multiple test tube (MTT) and Quanti-tray defined substrate MPN tests.

Recommendations for best use of H2Stest

Current recommendations for best application of the H2S test vary. Most studies agree the H2S test is a viable option when no other options exist in emergency or low-resource settings. Promoting the test as a motivational, educational, and empowerment tool on the community- and individual-level has also been advised (Sobsey & Pfaender 2002; Kaspar et al. 1992). Others have recommended conducting H2S testing in tandem with standard FIO methods (Gawthorne et al. 1996), comparing it to standard methods before widespread deployment (Yang et al. 2013a; Wright et al. 2012; Kaspar et al. 1992; Weppelmann et al. 2014a) or testing in conjunction with other inexpensive FIO field tests (Chuang et al. 2011). Nair et al. (2001) advocated for using the H2S test in developing countries where acceptable levels of fecal indicators in drinking water are <10 MPN/100 mL (Nair et al. 2001). Gawthorne et al. (1996) recommended using the H2S test as a presumptive test for Salmonella in drinking water in conjunction with coliform testing (Gawthorne et al. 1996).

(29)

19

rapidly, and affordably detect a variety of fecal indicator organisms makes studying the H2S test a worthy endeavor in the effort to provide microbiologically safe drinking water for all.

Compartment bag test method, format description, and history of validation

In 2007 researchers at the University of North Carolina at Chapel Hill and Duke University developed a simple kit for enumerating E. coli concentrations in water that is portable, relatively inexpensive, and provides easy-to-interpret results (Stauber et al. 2014). This kit, commonly referred to as the compartment bag test (CBT), is currently manufactured and distributed by Aquagenx, LLC (Chapel Hill, NC, USA). The CBT consists of a clear plastic multi-compartment bag into which 100 mL of water sample is distributed. Bacteria are detected using an E. coli growth medium containing a chromogenic substrate (5-bromo-4-chloro-3-indolyl-beta-D-glucuronic acid). The CBT does not require an incubator if ambient temperatures remain between 25-44.5° C. Positive results are indicated by a blue color-change in one or more of the compartments. Users can estimate E. coli concentrations by matching up the number and order of positive and negative compartments with a user-friendly table that follows a Poisson probability distribution assumption to generate discrete MPN/100 mL values and 95% confidence intervals (Appendix 1). Concentration estimates for the CBT are generated based on conventional MPN methods, calculating quantiles of the likelihood function of E. coli concentrations, and employing Bayesian Markov chain Monte Carlo (MCMC) analysis methods, as described by Gronewold et al. (2017). Bayesian analysis considers the probability distribution curves and likelihood functions of target microbes in samples to infer bacteria concentrations.

(30)

20

has grown in popularity as a user-friendly and cost-effective microbial detection kit and has been applied successfully in countries such as India, Vietnam, Nicaragua, Peru, Haiti, Ethiopia, Ghana, and Vanuatu (Weiss et al. 2016; Murcott et al. 2015; Stauber et al. 2014; Heitzinger et al. 2016; Adank et al. 2016; Morrison 2016; Gerges et al. 2016; McMahan et al. 2017).

Combining H2Sdetection medium with the compartment bag test

In a PhD dissertation McMahan (2011) studied the feasibility of a novel semi-quantitative H2S test which combined a commercial H2S detection substrate and culture medium with the compartment bag test. Several field and lab studies were conducted to compare the new test to traditional H2S and FIB detection methods. Comparisons included lab-made H2S medium vs. a commercial substrate

(PathoScreen™ by HACH, Loveland, CO), Whirl-pak plastic bags vs. plastic bottles, CBT versus Coliert Quanti-Tray to detect H2S-producing bacteria and E. coli, and H2S CBT vs. six popular field tests (MI Agar, BioRad Rapid E. coli 2 Agar, Coliert, Petrifilm E. coli, EasyGel, and E. coli CBT) for a cost per sample analysis. Overall, McMahan found the proprietary H2S powder to work as well as lab-prepared medium, the Whirl-pak bags to detect H2S on par with plastic bottles, and no significant difference between the CBT’s ability to detect H2S and Quanti-Tray’s ability to detect E. coli at similar incubation temperatures. The H2S CBT was also found to be the most cost-effective field test, at an estimated $0.40 USD per sample (other tests ranged from $1.70 - $15.00 USD per sample).

(31)

21

fecal bacteria and pathogens. They also found strong relationships of agreement between organisms identified by both methods tested. Researchers concluded the study provided an important step towards determining the H2S tests’ accuracy and specificity (McMahan et al. 2011).

Another study by McMahan et al. (2012) used biochemical (spread-plating with differential and selective agar) and molecular (TRFLP) methods to evaluate the ability of the H2S CBT test to associate with fecal indicator organisms, pathogens, and other microbes present in natural waters (lake, wells, and cistern rainwater) in North Carolina, United States. Researchers showed that water samples testing positive for H2S-producing bacteria also had bacteria of likely fecal origin and waters containing fecal pathogens were also positive for H2S bacteria. They also found that greater than 70% of isolates from natural waters were identified using TRFLP analysis and revealed a relatively stable group of organisms whose community composition differed with water source over time. The study further documented the validity of the H2S test for detecting and quantifying fecal contamination in water (McMahan et al. 2012).

Study niche and objectives

Since McMahan’s initial feasibility studies, no further work has been done to validate the H2S CBT as a new semi-quantitative H2S field test. The many promising aspects of the test, including its informative semi-quantitative format, affordability, ease-of-use, and longer shelf-life (approximately 3 years) compared to the E. coli CBT make the H2S CBT worth evaluating as a breakthrough microbial water quality test for emergency and low-resource environments. To better determine the reliability and quantification capabilities of the H2S CBT, the test must be compared with standard

(32)

22

(33)

23

CHAPTER 4: METHODS

The compartment bag test (CBT) and multiple test tube (MTT) technique were compared by collecting and testing lake water samples over the course of three months from July 2016 to September 2016 at the University of North Carolina at Chapel Hill in Chapel Hill, North Carolina, United States. Sample Sites

Surface water samples were collected from two reservoirs, University Lake (UL) and Cane Creek Reservoir (CC), that are used as municipal drinking water sources. University Lake is located in Chapel Hill, North Carolina and holds 450 million gallons of water and covers a surface area of 213 acres. The lake provides habitat for terrestrial and aquatic wildlife and is fed by five tributaries primarily passing through agricultural, suburban, and forested areas (Figure 1). Cane Creek Reservoir is located in Orange County, North Carolina and holds three billion gallons of water and covers a surface area of 540 acres. The lake is also a wildlife habitat and is fed by four tributaries passing primarily through agricultural and forested areas (Figure 2).

(34)

24

(35)

25

(36)

26 Sample Collection

Two replicate lake water samples were collected from each sampling site three times between July 2016 to September 2016, for a total of 60 field samples collected. Sampling alternated each week between University Lake and Cane Creek Reservoir so that each site was visited every other week.

Lake water samples were collected in autoclaved 1-L polypropylene containers using aseptic

techniques and sampling methods adhering to the Environmental Protection Agency’s (EPA) Region 4

surface water sampling methods (Decker & Simmons 2013). Upon collection, samples were stored in insulated containers filled with ice and transported immediately to a lab for processing. All samples were processed within six hours after field collection.

Physical and chemical environmental parameters were also collected at each sampling site during each sampling event using a YSI Professional Plus Multiparameter Instrument (Xylem Inc.). Parameters collected included air temperature (° C), water temperature (° C), pH, specific conductivity (μS/cm), and dissolved oxygen (% and mg/L). Current weather conditions and cloud cover (sunny, partially cloudy, and cloudy) were also recorded at each sampling site. In addition, total precipitation (cm) up to 72 h prior to sampling was collected for each sampling event using Weather Underground rain gauges stationed at the Horace Williams Airport in Chapel Hill, NC. The rain gauge is approximately three miles from University Lake and ten miles from Cane Creek Reservoir.

Sample Processing

(37)

S-27

producing bacteria cited in the literature. In total, 60 tests were conducted per sample dilution-temperature combination, resulting in 120 tests per method by dilution-temperature, 120 tests per method by dilution, and 240 tests total per method (Table 1).

The CBT method consisted of a clear polyethylene bag divided into five compartments of 1, 3, 10, 30 and 56 mL sample volumes (100 mL total) to allow for MPN enumeration. The method was

conducted using aseptic techniques adhering to the manufacturer’s user manual for drinking water

testing in the field (Aquagenx 2015). Briefly, 100 mL of diluted sample water was added to a sterile

plastic collection bottle. One HACH PathoScreen™ powder pillow was cut and poured into the collection

bottle and swirled to dissolve. Once dissolved, the sample solution was poured into each compartment of a sterile compartment bag test. The compartment bag was then sealed off to isolate the

compartments and placed in an incubator to promote bacterial growth.

The MTT method consisted of ten 16x150 mm glass test tubes from Fisher Scientific, each holding 10 mL sample volumes (100 mL total) to allow for MPN enumeration. The method was conducted using aseptic techniques adhering to the FDA’s Bacteriological Analytical Manual (Blodgett 2010). Briefly, 100 mL of diluted sample water was added to a sterile plastic collection bottle. One HACH

PathoScreen™ powder pillow was cut and poured into the collection bottle and swirled to dissolve. Once Table 1:Number of samples per CBT and MTT method based on dilution and temperature combinations.

Dilution Temperature (° C) Number of CBT Tests Number of MTT Tests Total

1:10 25 60 60 120

35 60 60 120

1:100 25 60 60 120

35 60 60 120

(38)

28

dissolved, the sample solution was added in 10 mL volumes to ten sterile glass test tubes. Tubes were capped and placed in an incubator to promote bacterial growth.

A set of positive and negative controls were included during each sampling event. The positive control (PC) consisted of a pair of CBT and MTT tests containing 100 mL sterile PBS inoculated with Salmonella enterica serovar Typhimurium strain LT2, a known H2S-producer. The negative control (NC) consisted of a pair of CBT and MTT tests containing only 100 mL sterile PBS media. PC and NC tests were diluted to 1:10, incubated at 35° C, and checked over the course of three days alongside field samples.

On days 1, 2, and 3 all tests were temporarily removed from incubators to check for growth. Any black liquid or solid color-change in compartment bags or test tubes indicated the presence of H2 S-producing bacteria. The combination of positive tubes or compartments for each test were recorded in a lab notebook and used to calculate MPN/100 mL and corresponding 95% confidence intervals (CI) as described below.

In addition to the CBT and MTT tests, 100 mL from each sample was processed through a Coliert-18 Quanti-Tray/2000 test (IDEXX Laboratories Inc., Westbrook, ME) to detect concentrations of total coliform and Escherichia coli (E. coli) in sample waters. The method was conducted using aseptic

techniques adhering to the manufacturer’s guidelines (IDEXX 2013). Samples were diluted 1:10 before processing and were incubated at 35° C for 18-20 h before checking for signs of bacterial growth. Results were computed as MPN/100 mL and reported with 95% CI.

Data Analysis

To obtain a single concentration estimate per sample, MPN results from each sample’s dilution

(1:10 and 1:100) were consolidated using the FDA-approved formula of Thomas (1942):

MPN/mL = (∑ gj) / (∑ tjmj ∑ (tj − gj) mj) (½)

(39)

29

Where the summation is over each dilution (1:10, 1:100) and ∑ gj denotes the number of positive tubes in the selected dilutions, ∑ tjmj denotes the grams of sample in all tubes in the selected dilutions, and ∑ (tj-gj) mj denotes the grams of sample in all negative tubes in the selected dilutions. To obtain approximate 95% confidence intervals, the FDA-approved method of Haldane (1939) was used to estimate the standard error of log10(MPN) of each sample:

Standard Error of Log10(MPN) = 1/(2.303 ∗ MPN ∗ (B^0.5))

Where B equals the sum of the exponents of each dilution’s negative MPN, multiplied by the dilution amount. Finally, 95% lower and upper confidence intervals were obtained using the following equation:

Log10(MPN) ± 1.96 ∗ (Standard Error

)

Applying the Thomas (1942) and Haldane (1939) equations resulted in 48 discrete MPN/100 mL and 95% CI outcomes for the CBT method, and 51 discrete MPN/100 mL and 95% CI outcomes for the MTT method (Appendix 3). The lowest level of detection for both tests was 9 MPN/100 mL, while the highest level of detection was 1812 MPN/100 mL. Discrete values were assigned to left- censored results to allow for inclusion in data analyses. Left-censored (all negative) results were assigned 4.5 MPN/100 mL, or half (½) the value of the lowest level of detection (9 MPN/mL). Right-censored (all positive) results were not included in data analyses, due to the inherent difficulty of determining true value of right-censored results according to Gronewold et al. (2017). A total of 10 right-censored results were removed, resulting in 350 observations of CBT-MTT paired tests over the course of three incubation days.

Figure 4: Part I of Method of Haldane (1939) equation

(40)

30

Data analysis was conducted using Excel 2016 and Statistical Analysis Systems (SAS 9.4). SAS data steps and proc steps were used to manage, transform, and compare paired CBT and MTT MPN/100 mL concentration estimates and 95% confidence intervals. Proc univariate was used to conduct Shapiro-Wilk normality tests on untransformed data and natural log (ln) and log base 10 (log10) transformed data. Descriptive statistics (proc means) were generated to compare CBT vs. MTT means, medians, and median differences and spearman’s correlation coefficients (proc corr) were run to test for direction and strength of method association. Contingency tables (proc freq) analyzed percent overlap of positive-negative results, risk-based categorical results, and acceptable drinking water range results. Frequency counts (proc freq) assessed percent overlap of CBT MPN estimates to paired MTT 95% upper and lower confidence intervals. CBT and MTT Log10(MPN) transformed values were compared to each other and the effect of days and temperature using a linear regression model (proc glm). A logistic regression model (proc logistic) was also used to evaluate the effects of incubation (temperature, days) and environmental (precipitation, air temperature, water temperature, pH, specific conductivity, and dissolved oxygen) conditions on the odds of CBT MPN values falling within paired MTT 95% CI.

Spearman’s correlation coefficients (proc corr) and linear regression (proc glm) were also used to compare replicate sample concentration estimates. Finally, spearman’s correlation coefficients (proc corr) were generated to observe relationships between CBT and MTT concentration estimations and total coliform and E. coli concentration estimations obtained from the IDEXX Coliert-18 Quanti-Tray/2000 test.

(41)

31

Sensitivity: percentage of water samples that have a presence of detectable H2S-producing bacteria that CBT correctly identifies as having H2S-producing bacteria

Specificity: percentage of water samples that have absence of detectable H2S-producing bacteria that CBT correctly identifies of having absence of H2S-producing bacteria

False Positive Rate: percentage of water samples that have absence of H2S-producing bacteria that CBT identifies having presence of detectable H2S-producing bacteria

False Negative Rate: percentage of water samples that have presence of detectable H2 S-producing bacteria that CBT identifies having absence of H2S-producing bacteria

Positive Predictive Value: percentage of water samples identified by the CBT as having presence of detectable H2S-producing bacteria that truly have presence of H2S-producing bacteria

Negative Predictive Value: percentage of water samples that the CBT identifies as having absence of detectable H2S-producing bacteria that truly have absence of H2S-producing bacteria • Accuracy: percentage of samples classified correctly by CBT

The 4x4 contingency table was generated by categorizing paired CBT and MTT results based on

the World Health Organization’s (WHO) health risk categories for E. coli in drinking water. Currently the WHO does not have similar categories for H2

S-producing bacteria, otherwise those definitions would

have been used. The WHO categorizes E. coli MPN/100 mL concentrations based on the following:

Safe: water sample contains less than 1 MPN/100 mL of E. coli bacteria

Intermediate Risk: water sample contains greater than 1 MPN/100 mL but less than or equal to 10 MPN/100 mL of E. coli bacteria

High Risk: water sample contains greater than 10 MPN/100 mL but less than or equal to 100 MPN/100 mL of E. coli bacteria

(42)

32

In addition, some developing countries accept drinking water with E. coli bacteria as long as levels are less than 10 MPN/100 mL. To account for regulations in these countries, a 2x2 contingency table comparing paired tests within these parameters was generated as well.

Quality control

(43)

33

CHAPTER 5: RESULTS

Overview of key findings

Several statistical comparisons were made between CBT and MTT results to determine the ability of the CBT to detect and quantify H2S-producing bacteria at similar rates and magnitudes to the MTT. A strong positive correlation was found between the CBT and MTT, particularly at 25° C incubation temperature over a 2 day period (44-48 h). Overview of descriptive statistics and linear regression analyses indicated the CBT consistently underestimated H2S concentrations as determined by the MTT. The odds that CBT MPN results would fall within paired MTT 95% CI bounds increased with a one unit increase in field water temperature (° C), and decreased with a one unit increase in field air temperature and pH. Additionally, a significant positive correlation was found between MTT H2S concentrations and

E. coli concentrations determined by the Coliert Quanti-Tray method. No significant relationship was found between the CBT and Quanti-Tray methods.

Normality tests for untransformed and transformed data

To determine if data were normally distributed, Shapiro-Wilk tests were applied to

(44)

34 Correlation coefficient analysis of method pairs

To test the direction and strength of association between the CBT and MTT methods, a Spearman Rank Correlation analysis was run on log10 transformed paired CBT-MTT MPN values. Correlation coefficients were generated comparing CBT and MTT results overall, by incubation

(45)

35

0.5 1.0 1.5 2.0 2.5 3.0

MTT Log10 (MPN/100 mL) 0.5 1.0 1.5 2.0 2.5 C B T L o g 1 0 ( M P N /1 0 0 m L ) Regression

Scatter plot of CBT (y-axis) and MTT (x-axis) Log10 (MPN/100 mL) values with best fit regression line

Temp=25 Day=2

Table 2: Spearman correlation coefficients (r)of compartment bag test (CBT) vs. multiple test tube (MTT) log10(MPN/100 mL) values for every combination of

incubation temperature and numbers of incubation days.

Temp (° C) Day N Obs Correlation (r) a

25

1 60 0.66

2 58 0.78

3 56 0.48

35

1 60 0.66

2 58 0.58

3 58 0.61

a All p-values highly significant (P<0.0001).

Figure 6: Scatter plot of the highest correlation (r = 0.78) between compartment bag test (CBT) log transformed (Log10) most probable number (MPN) per 100 mL values vs. paired MTT Log10(MPN/100 mL) values with best fit

regression line. Paired samples had been incubated for 2 days at 25° C.

CBT Lo g1 0 ( M P N /1 0 0 m L)

(46)

36

0.5 1.0 1.5 2.0

MTT Log10 (MPN/100 mL) 1.0 1.5 2.0 C B T L o g 1 0 ( M P N /1 0 0 m L ) Regression

Scatter plot of CBT (y-axis) and MTT (x-axis) Log10 (MPN/100 mL) values with best fit regression line

Temp=25 Day=1

Figure 7: Scatter plot of one of the higher correlations (r = 0.66) between compartment bag test (CBT) log transformed (Log10) most probable number (MPN) per 100 mL values vs. paired MTT Log10(MPN/100 mL) values

with best fit regression line. Paired samples had been incubated for 1 day at 25° C.

C

B

T

Log1

0

(M

PN

/10

0

mL

)

(47)

37

0.5 1.0 1.5 2.0 2.5 3.0

MTT Log10 (MPN/100 mL) 0.5 1.0 1.5 2.0 C B T L o g 1 0 ( M P N /1 0 0 m L ) Regression

Scatter plot of CBT (y-axis) and MTT (x-axis) Log10 (MPN/100 mL) values with best fit regression line

Temp=35 Day=1

Figure 8: Scatter plot of one of the higher correlations (r = 0.66) between compartment bag test (CBT) log transformed (Log10) most probable number (MPN) per 100 mL values vs. paired MTT Log10(MPN/100 mL) values

with best fit regression lin. Paired samples had been incubated for 1 day at 25° C.

C

B

T

Log1

0

(M

PN

/10

0

mL

)

(48)

38 Comparison of descriptive statistics

To compare similarities and differences in descriptive statistics for CBT and MTT concentration estimations, the mean, median, standard deviation, variance, and median difference for each method overall were calculated (Table 3). The mean, median, standard deviation, variance, and median difference were also calculated for every incubation time and temperature combination (Table 4). Finally, the same statistics were calculated for paired tests when H2S bacteria concentrations in samples were less than or equal to 100 MPN/100 mL (Table 5). To calculate median differences, a Wilcoxon Singed-Rank test was applied. All median difference results were statistically significant, indicating that there was a significant difference between CBT and MTT MPN medians. Regardless of incubation

0.5 1.0 1.5 2.0 2.5 3.0

MTT Log10 (MPN/100 mL) 1.0 1.5 2.0 2.5 C B T L o g 1 0 ( M P N /1 0 0 m L ) Regression

Scatter plot of CBT (y-axis) and MTT (x-axis) Log10 (MPN/100 mL) values with best fit regression line

Temp=25 Day=3

Figure 9: Scatter plot of the worst correlation (r=0.48) between compartment bag test (CBT) log transformed (Log10) most probable number (MPN) per 100 mL values vs. paired MTT Log10(MPN/100 mL) values with best fit

regression line. Paired samples had been incubated for 3 days at 25° C.

C

B

T

Log1

0

(M

PN

/10

0

mL

)

(49)

39

conditions, the mean and median for the CBT method were consistently lower than the corresponding MTT mean and median. Visual representations of paired tests were also generated though box & whisker plots showing log10 transformed CBT-MTT pairs overall (Figure 7), by incubation temperature (Figure 8), and by numbers of incubation days (Figure 9).

Table 3: Comparison of H2SMPN descriptive statistics for compartment bag test (CBT) vs. multiple test tube

(MTT) test overall, in terms of means, medians, standard deviations, variances, and median differences.

Test N Obs

Mean MPN/100 mL Median MPN/100 mL Std Dev MPN/100 mL Variance

(MPN/100 mL)2

Median Difference a

MPN/100 mL

CBT 350 76 27 116 13460 -224

MTT 350 300 106 436 190263

a Median difference determined by Wilcoxon Signed-Rank Test, results highly significant (P<0.0001).

Table 4: Comparison of H2SMPN descriptive statistics of compartment bag test (CBT) vs. multiple test tube

(MTT) test by incubation temperature and numbers of incubation days, in terms of means, medians, standard deviations, variance, and median differences.

Temp

(° C) Day Test N Obs

Mean MPN/100 mL Median MPN/100 mL Std Dev MPN/100 mL Variance

(MPN/100 mL)2

Median Difference a

MPN/100 mL

25 1 CBT MTT

60 19

42

5 b 10 30 71 882 4980 -23

2 CBT

MTT

58 72

345 28 108 103 487 10697 236899 -273

3 CBT

MTT

56 107 594 43 437 125 540 15555 291762 -488

35 1 CBT

MTT

60 29

164 9 43 52 316 2745 99577 -135

2 CBT MTT

58 76

320 39 170 118 392 13821 153398 -245

3 CBT MTT

58 160 359 115 170 159 452 25256 204035 -199

(50)

40

Test N Obs

Mean

MPN/100 mL

Median

MPN/100 mL

Std Dev

MPN/100 mL

Variance

(MPN/100 mL)2

Median Difference a

MPN/100 mL

CBT 164 17 5 b 21 439 -12

MTT 164 29 20 27 753

Table 5: Comparison of descriptive statistics (in terms of mean, median, standard deviation, variance, and median differences) for compartment bag test (CBT) vs. multiple test tube (MTT) test when H2S bacteria

concentrations in tests are less than or equal to 100 MPN/100 mL.

a Median difference determined by Wilcoxon Signed-Rank Test, results highly significant (P<0.0001) b This value represents left-censored data, rounded up from 4.5 to 5 MPN/100 mL.

Figure 10: Box & whisker plots of compartment bag test (CBT) log transformed (Log10) most probable number

(MPN) per 100 mL values vs. paired MTT Log10(MPN/100 mL) values overall (not separated by incubation

(51)

41

Figure 11: Box & whisker plots of compartment bag test (CBT) log transformed (Log10) most probable number

(52)

42 Presence/absence 2x2 contingency table by day

To determine how well the CBT detects H2S-producing bacteria compared to the MTT, a presence/absence (PA) 2x2 contingency table was generated for each CBT and MTT pair by incubation day (Table 6). From this table, CBT sensitivity, specificity, false positive rate, false negative rate, positive predictive value, negative predictive value, and accuracy were calculated (Table 7). A highly significant (P<0.0001) Fisher’s exact test and Cohen’s Kappa coefficient of 0.43 indicated a significant strength of agreement between CBT and MTT methods identifying PA results. The CBT demonstrated relatively high sensitivity (80%), specificity (86%), and positive predictive (98%) values, while displaying a very low false

Figure 12: Box & whisker plots of compartment bag test (CBT) log transformed (Log10) most probable number

(MPN) per 100 mL values vs. paired multiple test tube (MTT) Log10(MPN/100 mL) values based on numbers of

(53)

43

positive rate (2%). The CBT also demonstrated a high false negative rate (62%) and low negative predictive value (38%). The overall accuracy of the CBT was 81%.

Presence/absence 2x2 contingency table by sample

In addition to the previous 2x2 PA contingency table, paired CBT and MTT results were also compared by sample (Table 8). From this table, CBT sensitivity, specificity, false positive rate, false negative rate, positive predictive value, negative predictive value, and accuracy were calculated (Table 9). Results were not significant (P>0.05) due to the lack of negative results recorded by either test after 3 days of incubation. Although results were not significant, the false negative rate for the CBT decreased to 5% compared to 62% in Table 7.

Table 6: Presence/absence 2x2 contingency table of frequency and row percent overlap of the compartment bag test (CBT) vs. multiple test tube (MTT) method by incubation days.

Test Type a

CBT

MTT

Total

+

+

246

(98) b

6 (2) 252 (100) 61 (62) 37 (38) 98 (100)

Total 307 43 350

a Fisher’s exact tests highly significant (P<0.0001), Kappa coefficient 0.43.

b Numbers outside parenthesis denote frequency overlap, numbers inside parenthesis

denote row percent.

Table 7: Compartment bag test (CBT) sensitivity, specificity, false positive rate, false negative rate, positive predictive value, negative predictive value, and accuracy compared to multiple test tube method (MTT).

Test Sensitivity Specificity

False Positive Rate False Negative Rate Positive Predictive Value Negative Predictive Value Accuracy

(54)

44

2x2 contingency table using drinking water standards in developing settings by day

Most regulatory agencies focus on levels of fecal indicator bacteria in drinking water that are less than or equal to 100 MPN/100 mL sample. In addition, some developing countries consider the standard for safe drinking water to be when E. coli is at or below 10 MPN/100 mL sample. To determine how well the CBT performs within these standards, a 2x2 contingency table was generated comparing the ability of the CBT to MTT to classify samples between 1-10 MPN/100 mL and 11-100 MPN/100 mL of H2S-producing bacteria (Table 10). A highly significant Fisher’s exact test (P<0.0001) and kappa

coefficient of 0.53 showed a significant strength of agreement between the CBT and MTT tests identifying categorical results. The accuracy of the CBT under these standards was 76%.

Table 8: Presence/absence 2x2 contingency table of frequency and row percent overlap of the compartment bag test (CBT) vs. multiple test tube (MTT) method by sample.

Test Type a

CBT

MTT

Total

+

+

113

(99) b

1 (1) 114 (100) 6 (100) 0 (0) 6 (100)

Total 119 1 120

a Fisher’s exact test not significant (P>0.05)

b Numbers outside parenthesis denote frequency overlap, numbers inside parenthesis

denote row percent.

Table 9: Compartment bag test (CBT) sensitivity, specificity, false positive rate, false negative rate, positive predictive value, negative predictive value, and accuracy compared to multiple test tube method (MTT).

Test Sensitivity Specificity

False Positive Rate False Negative Rate Positive Predictive Value Negative Predictive Value Accuracy

CBT a 95% 0% 1% 5% 94% 0% 94%

(55)

45

2x2 contingency table using drinking water standards in developing settings by sample

In addition to the previous 2x2 contingency table, paired CBT and MTT results were also compared by sample comparing the CBT to MTT when H2S-producing bacteria concentrations were between 0-10 MPN/100 mL or 11-100 MPN/100 mL (Table 11). Results were not significant (P>0.05) due to the lack of test pairs recording results within 0-10 MPN/100 mL for the same sample. The accuracy of the CBT under these conditions was 77%.

Table 10: 2x2 contingency table of frequency and row percent overlap of the compartment bag test (CBT) vs. multiple test tube (MTT) based on concentrations of H2S-producing

bacteria between 1-10 MPN/100 mL and 11-100 MPN/100 mL by day.

Test Type a

CBT

MTT

Total

0-10 MPN/100 mL 11-100 MPN/100 mL

0-10 MPN/100 mL 65 (71) b

27 (29)

92 (100)

11-100 MPN/100 mL 12 (17)

60 (83)

72 (100)

Total 77 87 164

a Fisher’s exact tests highly significant (P<0.0001), Kappa coefficient 0.53.

b Numbers outside parenthesis denote frequency overlap, numbers inside parenthesis

denote row percent.

Table 11: 2x2 contingency table of frequency and row percent overlap of the compartment bag test (CBT) vs. multiple test tube (MTT) based on concentrations of H2S-producing

bacteria between 1-10 MPN/100 mL and 11-100 MPN/100 mL by sample.

Test Type a

CBT

MTT

Total

0-10 MPN/100 mL 11-100 MPN/100 mL

0-10 MPN/100 mL 0 (0) b

3 (100)

3 (100)

11-100 MPN/100 mL 2 (11)

17 (89)

19 (100)

Total 2 20 22

a Fisher’s exact test not significant (P>0.05).

b Numbers outside parenthesis denote frequency overlap, numbers inside parenthesis

Figure

Updating...

References

Updating...

Related subjects :